Jina AI Launches jina-reranker-v3.5: 0.6B Model Beats Qwen3-4B
JinaAI_ · x · 2026-08-03
Jina AI has released jina-reranker-v3.5, its latest 0.6B parameter listwise reranker. Building on v3, the new model is optimized for speed and tailored for enterprise data search.
It achieves an 63.20 nDCG@10 on the BEIR benchmark, outperforming Qwen3-Reranker-4B with approximately 7× fewer parameters. Across four benchmarks, the model sits on the Pareto front, meaning no evaluated model is both smaller and better.
Related event: Jina AI Launches jina-reranker-v3.5(2 posts)→
More from Models
- DeepSeek Models Tested: 242 One-shot Outputs Compared — kms_dev · 2026-08-03
- MiniMax H3 Tested: Users Report Censored Video Generation — DaLyon92x · 2026-08-03
- Google AI Mode Becomes Rigid Due to Over-Optimization for Safety and Shopping — Build-your-own-2020 · 2026-08-03
- AI Educator Teams Up with Hugging Face for Open Model Series — HarperSCarroll · 2026-08-03
- Anthropic Investigating Degraded Performance Issue on Claude Sonnet 5 — ClaudeAI-mod-bot · 2026-08-03
- User extracts internal system prompt of Google Gemini — Empty_Paramedic_5957 · 2026-08-03